3385 results about "Computation complexity" patented technology

The invention belongs to the field of natural language processing, and provides a task complexity driven graph semantic multi-agent collaborative decision-making method and system.The task complexity driven graph semantic multi-agent collaborative decision-making method comprises the steps that a task text is obtained and subjected to semantic coding to obtain a task semantic vector, evaluation is conducted based on the task semantic vector to obtain a complexity vector, and a task complexity score of the complexity vector is calculated; the task semantic vector and the complexity vector are fused to obtain a task representation vector, an agent capability relation graph is constructed, the participation probability of each agent node is obtained according to the task representation vector and the agent capability relation graph, and a dynamic agent combination scheme is formed; and performing task decomposition according to the agent combination scheme, constructing a sub-task dependency graph, scheduling the execution sequence of the sub-tasks through topological sorting, realizing cooperative execution of the agents, and generating a task result. According to the method, precise matching and efficient cooperation of the agent combination are realized, and the capability of processing complex tasks and the resource utilization efficiency of the multi-agent system are remarkably improved.

The invention relates to the technical field of artificial intelligence and natural language processing, in particular to a key value cache compression and sparse attention calculation method and system for large language model reasoning, and the method comprises the steps: an offline calibration stage; the online reasoning stage comprises the following steps: a pre-filling step; an autoregression generation step: for each newly generated lexical element, projecting a current query vector Q and a key vector K in a key cache to a low-dimensional space to obtain Q'and K '; calculating an approximate attention score based on Q'and K ', and selecting an index I of the first k most relevant lexical elements which are ranked from high to low; and calculating an accurate attention score based on Q and K [I], and calculating with the value vector V [I] to obtain the output of the current lexical element. According to the scheme, the memory and calculation bottleneck of large model reasoning in a scene of long text sequence input are solved, and the method has the advantages of reducing video memory occupation and calculation complexity at the same time.

Deep learning tools such as convolutional neural networks (CNNs) and transformers have spurred great advancements in computational biology. However, existing methods are constrained architecturally in context length, computational complexity, and model size. This application introduces a sub-quadratic architecture for modeling, which combines projected gated convolutions and structured state spaces to achieve local and global context with, for example, single-nucleotide resolution. These models outperform CNN-, GPT-, BERT-, and long convolution-based models in many tested genomics tasks without pre-training and with 4×-781× fewer parameters. In the proteomics domain, these models similarly outperform pretrained attention-based models, including ESM-1B and TAPE-BERT, on remote homology prediction without pre-training and while using 3,308×-23,636× fewer parameters.

The invention provides an image restoration and super-resolution reconstruction system and method based on deep learning, and belongs to the technical field of digital image processing. The invention aims to solve the problems of high calculation complexity and resource consumption, limitation of long sequence processing, high training difficulty and texture scene deficiency when a multi-scale residual network based on a Transform architecture is used for image resolution conversion. The reconstruction system comprises: an image preprocessing module performing window division and video memory optimization on an input low-resolution image; the multi-layer fusion network dynamically adjusts the characteristics of the low-resolution image, captures channel information in different scenes, performs interactive fusion, performs comparison supervision, establishes an information communication channel, dynamically adjusts and optimizes parameters through negative feedback, and obtains a super-resolution image. And the loss function module maximizes the similarity of the super-resolution image and the high-resolution image in the segmentation feature space to obtain a final super-resolution image.

The invention provides an intelligent boxing method and system. The intelligent boxing method comprises the steps of obtaining boxing task information and generating a boxing scheme. The boxing task information comprises the size, the weight, the stacking limitation, the number and the priority of each cargo, and the size, the bearing capacity and the occupation state of each loading unit. And under the condition that the size, bearing and stacking constraint conditions are met, a plurality of candidate boxing combinations are generated, the loading utilization rate, the residual space distribution and the shape matching degree are evaluated based on a preset optimization target, and an optimal scheme is selected. The system comprises an input terminal, a task information receiving module, a data storage module, a task and scheme storage module, an operation processing module and a final boxing scheme generation module. According to the method, multi-dimensional constraint matching and weight balance are considered, and space waste and transportation risks are avoided; the maximum greedy algorithm and the adaptive large neighborhood search optimization are combined to realize the combination of automation and experience. The method can be expanded to batch task scheduling and grouping sequencing, so that the calculation complexity is reduced, and the overall boxing and transportation efficiency is improved.

The invention discloses a virtual power plant global optimization scheduling method, system and device based on cloud edge collaboration, and a storage medium, and belongs to the technical field of power system scheduling. The method comprises the following steps: based on a cloud edge coordinated regulation and control framework comprising a cloud layer, an edge layer and an end side layer, taking minimization of the total operation cost of a system as a target, comprehensively considering a power balance constraint, a main network interaction constraint, a distribution network transmission constraint, a distributed resource operation constraint, an energy storage equipment constraint and a renewable energy consumption constraint; establishing a global optimization scheduling model; edge collaborative optimization is realized by adopting an alternating direction multiplier method, a global coupling problem is decomposed into local optimization sub-problems and a cloud coordination problem of each region, and aggregation power information is sent to the cloud after the local optimization sub-problems are solved in parallel in each region; and the cloud performs global coordination optimization to generate an optimal scheduling strategy, and issues a scheduling instruction to the edge layer to control the actual operation of the distributed power supply, the energy storage equipment and the controllable load, thereby realizing the collaborative optimization scheduling of the virtual power plant. The problems that in virtual power plant large-scale distributed resource coordination optimization, calculation complexity is high, communication burden is heavy, and real-time performance and global optimality are difficult to consider at the same time are effectively solved.

The invention discloses a power transmission line foreign matter detection method and system based on multi-modal image fusion, and relates to the technical field of intelligent operation and maintenance and state monitoring of a power system, a lightweight Ev-Mama architecture is introduced into a backbone network part of YOLOv13, the model keeps relatively low calculation complexity, and meanwhile, the power transmission line foreign matter detection efficiency is improved. And the modeling capability of the method on the long-range dependency relationship and the global semantic information is obviously enhanced. Besides, by using the CDIDF module, the EVCS module and the MHSAA module, on the basis of increasing a small amount of calculation, the scale sensing ability, the space structure modeling ability and the context understanding ability of the model are effectively improved, and the performance bottleneck of a traditional YOLO series network in the aspects of processing small targets, shielding targets and cross-scale information fusion is effectively relieved.

The invention discloses a software development task intelligent allocation method and system based on a multi-dimensional capability portrait, and the method comprises the following steps: 1, obtaining multi-source development data generated by a developer in a development process and demand description data of a to-be-allocated software development task, and carrying out the preprocessing, and forming a structured data set; according to the method, a multi-dimensional ability portrait covering technical ability, project experience, collaboration attributes and performance is constructed, a privacy-protected distributed learning mechanism is adopted for dynamic updating, dominant and implicit requirements of tasks are analyzed in combination with natural language processing, complexity and dependency are calculated, and the performance of the performance is improved. A dynamic task feature vector corresponding to a capability feature vector dimension is constructed, and meanwhile, a self-adaptive adjustment mechanism based on real-time data monitoring and online learning is designed to form data closed-loop feedback, so that the problems of one-sided capability evaluation, staticizing task demand analysis and lack of the self-adaptive adjustment mechanism are comprehensively solved; and accurate and intelligent distribution of software development tasks is realized.

The invention relates to the technical field of image recognition, in particular to a lightning arrester fault thermal imaging picture recognition method and system combined with gaze prediction, and the method comprises the steps: carrying out the down-sampling of an original thermal imaging image to a fixed size, sequentially passing through a multi-layer convolution and a Spatial Softmax layer, and outputting a predicted gaze point track sequence, extracting a key area and a non-key area according to the intensity of the fixation point track sequence; reconstructing the key area to obtain a compressed and recombined target area; inputting the non-key region into a variational auto-encoder to obtain a low-dimensional potential feature vector; and inputting the image-structure joint feature representation into an EffiCroprViT model, and finally obtaining a lightning arrester fault classification result. Local and global features are efficiently fused, the calculation complexity is reduced, background noise interference is effectively suppressed, real-time and accurate identification and early warning of the fault state of power equipment are realized, and the fault classification method has the advantages of high efficiency, high reliability and high reliability. Therefore, the safety and stability of power grid operation are ensured.

The invention discloses a method for improving long text processing efficiency and accuracy, and relates to the technical field of natural language processing and large language models.According to the method, text word segmentation embedding, sliding block preprocessing, YaRN position code injection, dynamic sparse attention calculation, multi-level attention fusion, graded KV cache management and output generation are sequentially executed; position drift is inhibited through logarithmic scaling, and key contexts are adaptively screened according to the attention activeness, so that the attention calculation complexity is close to linearity; in million-level Token reasoning, the video memory occupation of the method is reduced, the remote dependency recall rate is improved, and the method is suitable for scenes such as document analysis, code auditing and multi-mode streaming understanding.

The invention relates to the field of artificial intelligence medical application, and discloses a lightweight intelligent traditional Chinese medicine inquiry system and a construction method thereof, and the system comprises a multi-dialect adaptive speech recognition module, a traditional Chinese medicine intelligent dialogue large language model module, a natural speech synthesis module, and a continuous learning mechanism module. The multi-dialect adaptive speech recognition module is used for converting dialect speech input of a patient into a standard text; the traditional Chinese medicine intelligent dialogue big language model module is the core of the system and is used for carrying out natural language understanding, dialectical reasoning and inquiry dialogue generation, and the natural speech synthesis module is used for converting a text response generated by the system into speech output; and the continuous learning mechanism module realizes continuous optimization of the large language model through incremental learning architecture and clinical feedback integration. According to the method, while the professional traditional Chinese medicine diagnosis capability is maintained, the calculation complexity is remarkably reduced, and the universality and sustainable development capability of system application are improved.

The invention relates to a hyperspectral image classification method and device based on a state space model. The hyperspectral image classification method based on the state space model comprises the following steps: sequentially carrying out feature extraction and serialization processing on hyperspectral image data to obtain a shallow feature projection vector; performing global-local feature extraction on the shallow feature projection vector by adopting a neural network based on a state space model to obtain a fused feature projection vector; and carrying out pixel-by-pixel classification and dimension rearrangement on the hyperspectral image data in sequence to generate a classification result of the hyperspectral image data. According to the hyperspectral image classification method based on the state space model, long-range dependence modeling is achieved through the neural network based on the state space model with linear complexity, the calculation complexity is effectively reduced, and through feature fusion and residual error connection, the classification accuracy of the hyperspectral image is improved. And the perception capability of the neural network on different scale space-spectrum structures in the hyperspectral image is effectively enhanced.

The invention relates to the technical field of biomedical engineering and computer vision, in particular to a non-contact physiological signal extraction method and system based on frequency self-adaption and illumination noise perception.The method comprises the following steps of multi-mode video stream collection and spatio-temporal data preprocessing, illumination-noise perception mask generation and feature filtering, multi-mode video stream collection and spatio-temporal data preprocessing, illumination-noise perception mask generation and feature filtering, and non-contact physiological signal extraction. Frequency adaptive gating and frequency domain feature enhancement, depth time attention feature re-calibration, physiological signal regression and closed loop optimization; the method has the beneficial effects that a lightweight end-to-end deep learning network architecture is constructed by systematically fusing three core modules of illumination-noise perception mask, frequency adaptive gating and depth time attention, and the defects that a traditional physical model depends on artificial prior and is poor in anti-interference performance and high in reliability are overcome. And the one-sidedness caused by high calculation complexity and difficulty in distinguishing the signal and noise of the existing deep learning model is avoided, and the weak physiological signal can be recovered from the face video more accurately and robustly.

The invention discloses a multi-mode emotion continuous recognition method for medical treatment, belongs to the technical field of artificial intelligence and medical treatment information, and mainly aims to simulate the dynamic change process of emotion by establishing a Neural ODEs framework and overcome the static property and discreteness of emotion modeling in a traditional method. Through a causal inference technology, emotional features are separated from individual-independent physiological differences, and the generalization ability across individuals is improved. A self-supervised learning method is utilized, the synergistic effect between the EEG and the eye movement signal is improved through cross-modal contrast learning, and the emotion recognition precision is enhanced. The calculation complexity is reduced through a dynamic sparse attention mechanism, and meanwhile, focusing is performed on a key time slice in emotion recognition. Through multi-task joint learning, the model learns multiple tasks such as emotion intensity regression and tested identity recognition during emotion classification, and the personalized emotion recognition capability is improved.

The invention discloses a lightweight multi-source unmanned aerial vehicle target detection method and system based on a DEYOLO framework, and relates to the field of target detection, and the method comprises the steps: obtaining an unmanned aerial vehicle visible light image and an unmanned aerial vehicle infrared image which are registered, and inputting the images into a pre-trained target detection model; the model comprises a double-flow feature extraction network module which is used for extracting an unmanned aerial vehicle visible light image and an unmanned aerial vehicle infrared image to obtain a visible light feature map and an infrared feature map; the bimodal adaptive feature weighting module is used for performing bimodal adaptive feature weighting and adding on the visible light feature pattern and the infrared feature pattern to obtain fusion features; the lightweight bimodal attention enhancement module is used for performing feature enhancement on the fusion features; and the detection head is used for detecting the enhanced features. According to the method, the calculation complexity is effectively reduced, and the detection precision and the reasoning speed of the model on the low-slow small target and the robustness of the model on a complex scene are remarkably improved.

The invention discloses a photovoltaic energy storage cooperative control method and system, and the method comprises the steps: obtaining the real-time state data and environment information of distributed photovoltaic and energy storage equipment, and constructing a system network topology; constructing a hierarchical prediction model including photovoltaic power generation prediction, load demand prediction and electricity price prediction based on the system state data set; constructing a multi-objective function containing an economic benefit objective and a system stability objective based on a prediction result, and performing objective decomposition through recursive iteration; according to a local optimization target, generating a photovoltaic power adjustment strategy and an energy storage charging and discharging strategy by reducing a distributed maximum matching gap and reducing an order to break a distributed maximum independent set obstacle algorithm; and based on the distributed control strategy set, the system operation state is monitored in real time, and control parameters are dynamically adjusted by adopting an event triggering mechanism. According to the method, the problems of calculation complexity and communication dependence of traditional centralized control in a large-scale photovoltaic energy storage system are solved.

The invention provides a microscopic automatic focusing method and system based on image gray histogram characteristics, and the method comprises the steps: collecting an image sequence under different focal lengths, carrying out the fuzzy processing, extracting a gray histogram of each frame of image, and calculating the peak position and full width at half maximum of the histogram as the evaluation characteristics of the image definition; calculating the variance of each feature and automatically allocating a weight according to the relative response degree; and finally, comprehensively evaluating the image definition through a weighted definition scoring function, and selecting the focal length corresponding to the image with the optimal score as the optimal focusing position. The method is based on the global features of the gray histogram, is high in anti-noise capability, is adaptive to different imaging scenes through weight adaptive adjustment, is low in calculation complexity, supports real-time focusing, is especially suitable for high-noise fluorescence microscopic imaging scenes, is high in system portability, is low in operation threshold, and effectively improves the accuracy and stability of microscopic automatic focusing.

A hybrid neural network architecture is disclosed that integrates matrix multiplication-free (MatMul-free) transformation layers with spiking neural network (SNN) layers for efficient, low-power computation. The system includes an interface module configured to convert intermediate continuous-valued data from MatMul-free layers into a spike-compatible format using encoding techniques such as rate coding, phase coding, or threshold-based conversion. The SNN layers process the spike-encoded data in an event-driven manner, enabling sparse, temporal inference. Training is supported by a hybrid optimization strategy combining backpropagation in MatMul-free components with surrogate gradient descent or spike-timing-dependent plasticity (STDP) in SNN layers. The architecture reduces computational complexity, supports real-time adaptability, and enables deployment in energy-constrained environments such as edge devices and neuromorphic platforms. The system may be implemented in hardware, software, or a co-designed pipeline optimized for dynamic sensor data, control signals, or continuous inference tasks.

The invention relates to the technical field of quantum computing, and discloses a quantum circuit mapping method and system based on deep reinforcement learning, according to the method, an original logic quantum circuit is analyzed and simplified into a ZX-graph, then the ZX-graph is converted into a quantum circuit dependency graph, the computing complexity is remarkably reduced, and the quantum circuit mapping efficiency is improved. A large-scale quantum circuit mapping task can be processed within reasonable time, a Markov decision process environment is constructed through a quantum circuit dependency graph and quantum chip parameters, a deep reinforcement learning agent is trained by using the Markov decision process environment, and an optimal mapping strategy optimization model is obtained. Through the optimal mapping strategy optimization model, the corresponding optimal quantum bit mapping scheme can be output to the current to-be-mapped logic line, so that the method can be adapted to quantum chips of various topological structures, and the generalization ability and adaptability of the mapping scheme are improved.

The invention relates to an Internet of Vehicles multi-agent cooperative sensing method and system. The method comprises the following steps: constructing a collaborative sensing network, wherein the collaborative sensing network comprises a self-agent and a plurality of collaborative agents; acquiring and processing sensing data through the collaborative sensing network; performing feature extraction to obtain intermediate features; self-adaptive sparsification is carried out to obtain sparse features, and the sparse features are compressed and transmitted to a self-agent; performing time sequence feature enhancement on the features of all the agents at the self-agent end; fusing the features to obtain fused features; and constructing an Internet of Vehicles perception model, and realizing perception by the detection model according to the fused features. According to the method, the calculation complexity of traditional global attention is reduced from the square level to the linear level through an adaptive sparsification mechanism, the calculation overhead is remarkably reduced while the multi-agent feature interaction precision is kept, and the method is more suitable for real-time operation on the vehicle-mounted edge equipment with limited resources.

The invention discloses an Internet of Vehicles intrusion detection method based on hierarchical feature enhancement and a dynamic sparse attention mechanism, and the method comprises the steps: firstly obtaining an original CAN bus message data stream, and carrying out the multi-modal data preprocessing and hierarchical feature construction; secondly, for hierarchical features, performing message-level depth feature compression and space-time statistical feature enhancement by adopting a dual-channel feature compression mechanism to form an enhanced mixed feature vector, and then constructing a neural network model of a dynamic sparse attention mechanism based on content awareness to perform feature extraction; and inputting a sparse attention calculation result into a hierarchical classifier, and predicting and outputting a detection result. And training the network model, and deploying the trained network model in a vehicle-mounted embedded device to realize real-time message monitoring and intrusion detection. According to the method, the calculation complexity and parameter quantity of the model are greatly reduced, the real-time performance in a vehicle-mounted environment is remarkably improved, and the detection capability on complex hidden attacks is effectively improved.

The invention relates to the technical field of large model algorithms, in particular to a large model dynamic compression optimization method and system based on sparse pruning, and the method comprises the steps: capturing original weight fluctuation data generated by resource fluctuation in reasoning, and obtaining sparse weight reference data through sparse processing; analyzing calculation complexity through model reasoning delay data, and separating reasoning delay amount caused by a model scale; dynamically controlling the model compression ratio within a preset performance range based on the delay amount and the sparse reference data, and collecting reasoning precision distribution data under different compression parameters; evaluating a model performance state under each parameter by means of a neural network simulation method, and generating a performance state simulation result; determining a model quality optimization compensation parameter based on a simulation result by combining resource fluctuation data acquired in real time in a compression process; and finally, the compression strategy is adaptively regulated and controlled through the compensation parameters, and collaborative optimization of model calculation complexity, reasoning precision and delay during dynamic change of hardware resources is realized.

The invention provides a camera external parameter calibration method and device based on image and point cloud matching, and the technical scheme of the invention is that a point cloud rendering view corresponding to an original point cloud is rotated, so that the point cloud rendering view and a camera image are superposed visually, and initial space association is established; and directly learning and solving accurate camera external parameters from the local relevance between the original point cloud intensity information and the image RGB information in an end-to-end manner by using a deep learning model. According to the scheme, the strict camera-radar orientation consistency requirement in a traditional method is not needed through rough matching, manual participation in the whole process of traditional calibration is replaced, transition from visual alignment to geometric alignment is achieved, and the calibration efficiency and scene adaptability are remarkably improved. Meanwhile, the scheme of the invention initiates a process of visual field cone cutting-perspective projection rasterization-intensity-RGB end-to-end matching, and compared with a traditional feature point method, the calculation complexity is greatly reduced, and the memory occupation is reduced by 80%.

The invention provides a computing power network dynamic topology modeling method and system, electronic equipment and a storage medium, and aims to solve the problems of large dependent data volume, high calculation complexity and high model complexity of computing power network dynamic topology modeling. The method comprises the following steps: collecting multi-dimensional indexes of computing power nodes in real time; constructing a multi-dimensional index causal graph model of computing power nodes based on the collected multi-dimensional indexes; the influence of topological change on task scheduling is predicted through a causal graph model, when negative influence is predicted, a topological adjustment scheme is automatically generated, and a topological reconstruction strategy is generated; and simulating the generated topology reconstruction strategy, and verifying the validity of the strategy in the current fault scene. According to the invention, collaborative optimization and intelligent scheduling of computing power resources and network resources can be realized to cope with dynamic, heterogeneous and large-scale challenges in a computing power network.

The invention discloses a power distribution network transient characteristic prediction method based on supervised learning, and relates to the technical field of power distribution network state prediction, and the method comprises the steps: collecting historical operation data through a power distribution network monitoring system, carrying out the data preprocessing, and obtaining standardized multi-dimensional time series data; carrying out transient feature extraction, constructing a high-dimensional feature set, and carrying out feature dimension reduction according to a transient event tag to generate a feature subset; inputting the feature subset into a mixed supervised learning model of a gradient boosting decision tree GBDT and a long short-term memory network LSTM for joint training to obtain a transient feature prediction result; and calculating a root-mean-square error according to the transient characteristic prediction result and the real-time monitoring observation value of the power distribution network, and dynamically adjusting hyper-parameters of the supervised learning model based on a Bayesian optimization algorithm. According to the method, the detection accuracy can be improved, the calculation complexity can be reduced, and the discrimination capability and the time sequence prediction capability of the model are considered.

The invention discloses a high-dimensional data feature selection method and system based on a multi-strategy improved whale optimization algorithm, and the method guarantees the uniform distribution of populations through a good point set initialization strategy, and solves a search blind area problem caused by conventional random initialization. A whale optimization and particle swarm optimization double-population cooperation mechanism is adopted, and dynamic balance of global exploration and local development is achieved; and a tangential flight disturbance strategy is introduced, so that the capability of jumping out of local optimum of the algorithm is effectively enhanced. Finally, binary feature selection vectors are output and directly applied to machine learning model training, the classification precision is remarkably improved in the fields of medical diagnosis, image recognition and the like, the calculation complexity is reduced, and an efficient and reliable solution is provided for high-dimensional data feature selection.

The invention provides a lithium ion battery health state lightweight detection method based on a physical information neural network, and the method comprises the steps: collecting the time, voltage, current, temperature and state-of-charge data of a battery in a takeoff and landing stage discharge process, processing the data into takeoff and landing stage discharge time sequence data, and carrying out the detection of the lithium ion battery health state based on the takeoff and landing stage discharge time sequence data. The method comprises the following steps: designing characteristic factors related to battery aging, screening the characteristic factors by utilizing a Pearson's correlation coefficient and a grey relational degree algorithm to obtain optimal characteristic sequence data, inputting the optimal characteristic sequence data into a physical information neural network model constructed by two serially connected neural networks for training, and in the training process, obtaining the optimal characteristic sequence data. And performing hyper-parameter tuning on the two neural networks by adopting a Bayesian optimization algorithm, then performing fine tuning on the second neural network by adopting a hierarchical transfer learning strategy, and finally applying the trained physical information neural network model to battery health state detection. The method improves the quality of feature data, reduces the calculation complexity of features and models, and achieves the accuracy and reliability of the detection of the health state of the battery under the airborne condition.

The invention belongs to the technical field of remote sensing image analysis, and provides a multi-source remote sensing building extraction method based on hybrid experts, electronic equipment and a storage medium. The method comprises the steps of basic model construction, two-way structure model construction, optical feature extraction, earth surface feature extraction, two-way feature fusion, ViT coding, feature injection and extraction and semantic alignment step-by-step up-sampling. According to the invention, a dual-path structure is adopted to assist the multi-scale encoder and the bidirectional attention fusion module, and DSM data and optical images are used for feature fusion, so that the capability of distinguishing buildings from backgrounds is improved; the hybrid expert LoRA structure is introduced into the ViT encoder, parameters of the feedforward neural network are dynamically adjusted, the adaptability and flexibility of the model to input features are further enhanced, and the calculation complexity is reduced.

The invention discloses a low-light image enhancement method combining a state space model and wavelet transform, and aims to solve the problems of detail loss, inaccurate illumination estimation and high calculation complexity of the existing low-light enhancement algorithm. According to the method, illumination estimation, frequency domain enhancement, wavelet transform, a high-frequency attention mechanism, dynamic state space modeling (Mamba) and a multi-scale U-Net structure are introduced, so that high-quality and detail-retaining enhancement of a low-illumination image is realized. According to the method, firstly, based on the Retinex theory, an illumination prior image is generated by calculating the global mean value of RGB channels of a low-illumination image, illumination features are extracted in combination with an illumination estimation module, and a preliminary enhanced image is generated; then, a multi-scale dynamic coding and decoding network is adopted, a coding layer compresses spatial dimensions step by step and increases feature channels, a decoding layer recovers resolution step by step to preliminarily enhance image fusion, and an enhancement result is well obtained.

The invention discloses an efficient medical image segmentation method considering global modeling and local enhancement, and relates to the technical field of image segmentation. According to the method, adaptive space shift operation is executed in different directions through the AS-MLP module, the long-range dependence modeling capability is effectively enhanced, and the recognition performance of a complex structure focus is improved; the channel and space double attention mechanism and multi-scale convolution of the LMCAM module are combined, so that fine-grained feature extraction is realized, and the segmentation precision of the lesion boundary and the small-scale structure is remarkably improved; a lightweight network design is adopted, the calculation complexity is low, the reasoning speed is high, and the method is suitable for resource-limited clinical terminals and real-time diagnosis application; besides, the method has good cross-modal adaptability, can keep stable and efficient segmentation performance in various medical imaging modalities such as CT, MRI, ultrasound and dermatoscope, and has wide application value.